Meta’s Pyrefly Python type checker published a new AI policy and expanded contributing guidelines after coding agents flooded the project with pull requests and issues that take longer to review than to generate. The policy’s central tenet is “we want to interact with you, not your AI”: it bans LLM-generated communication, autonomous-agent contributions and pull requests not thoroughly reviewed by their human author, and blocks AI from one-shotting issues labeled “good first issue”, while still allowing AI for learning, investigation, coding and validation as long as contributors understand and own the result. The maintainers describe a growing backlog of untriaged issues, review-nudge traffic and burnout risk, and argue that issue and PR volume and throughput no longer signal a project’s health.
Pyrefly tightens AI contribution policy as agents strain open source review
Meta's Pyrefly Python type checker published a new AI policy and expanded contributing guidelines after coding agents flooded the project with pull requests and issues that take longer to review than to generate. The policy's central tenet is "we want to interact with you, not your AI": it bans LLM-generated communication, autonomous-agent contributions and pull requests not thoroughly reviewed by their human author, and blocks AI from one-shotting issues labeled "good first issue", while still allowing AI for learning, investigation, coding and validation as long as contributors understand and own the result. The maintainers describe a growing backlog of untriaged issues, review-nudge traffic and burnout risk, and argue that issue and PR volume and throughput no longer signal a project's health.
Source: Pyrefly